AODV-HM: A Hybrid Mesh Ad-hoc On-demand Distance Vector ...

Journal of Research and Practice in Information Technology, Vol. 41, No. 1, February 2009 65

AODV-HM: A Hybrid Mesh Ad-hoc On-demand DistanceVector Routing ProtocolAsad Amir PirzadaNICTA, Queensland Research Laboratory, 300 Adelaide Street, Brisbane, QLD 4000, Australia

Marius Portmann and Jadwiga IndulskaSchool of ITEE, The University of Queensland, Brisbane, QLD 4072, Australia{Asad.Pirzada, Marius.Portmann, Jadwiga.Indulska}@nicta.com.au

Wireless Mesh Networks (WMNs) have recently gained increasing attention and have emerged as atechnology with great potential for a wide range of applications. WMNs can be considered as asuperset of traditional mobile ad-hoc networks (MANETs), where the network is comprised ofmobile client devices MESH_CLIENTs. In addition to MESH_CLIENTs, a WMN can also containrelatively static devices called mesh routers (MESH_ROUTERs). Such hybrid WMN arecharacterized by a high level of heterogeneity, since static MESH_ROUTERs are typically muchless resource constrained than mobile MESH_CLIENTs, and are also often equipped with multipleradio interfaces. Traditional ad-hoc routing protocols do not differentiate between these types ofnodes and therefore cannot achieve optimal performance in hybrid WMNs. In this paper, wepropose simple extensions to the Ad-hoc On-demand Distance Vector (AODV) routing protocol,which aim to take advantage of the heterogeneity in hybrid WMNs by preferentially routing packetsvia paths consisting of high capacity MESH_ROUTERs. In addition, we implement a simplechannel selection scheme that reduces interference and maximizes channel diversity in multi-radioWMNs. Our simulation results show that in hybrid WMNs, our extensions result in significantperformance gains over the standard AODV protocol1.

Classification: C2.2 [Computer-Communication Networks]: Network Protocols — Routingprotocols

General Terms: Mesh, Routing, PerformanceAdditional Key Words and Phrases: Wireless Mesh Networks, Routing, Hybrid Mesh Networks,

Channel Diversity

Manuscript received: 15 April 2008Communicating Editor: Colin Fidge

Copyright© 2009, Australian Computer Society Inc. General permission to republish, but not for profit, all or part of thismaterial is granted, provided that the JRPIT copyright notice is given and that reference is made to the publication, to itsdate of issue, and to the fact that reprinting privileges were granted by permission of the Australian Computer Society Inc.

1. INTRODUCTIONWireless Mesh Networks (WMNs) are self-organizing and self-configuring wireless networks,typically implemented with IEEE 802.11 hardware. In conventional wireless LANs, clientscommunicate with access points via a single-hop wireless link and access points are interconnectedvia a wired backbone infrastructure. WMNs do not rely on such a wired backhaul and implementconnectivity via a wireless multi-hop network. Their robustness, self-organizing and self-

1 This paper extends upon our previous research paper titled “Hybrid Mesh Ad-hoc On-demand Distance Vector Routing Protocol”, pub lishedin the proceeding of the Thirtieth Australasian Computer Science Conference (ACSC'07), Ballarat, Australia, Vol 29(1), pages 49-58.

JRPIT 41.1.QXP:Layout 1 13/03/09 12:58 PM Page 65

AODV-HM: A Hybrid Mesh Ad-hoc On-demand Distance Vector Routing Protocol

Journal of Research and Practice in Information Technology, Vol. 41, No. 1, February 200966

configuring nature, and the low cost of wide area deployment make WMNs an attractive platformfor a wide range of applications, such as public safety and emergency response communications,intelligent transportation systems, or community networks.

We can differentiate between two types of nodes in a WMN: MESH_ROUTERs andMESH_CLIENTs. MESH_ROUTERs are relatively powerful and static nodes, which have eitheraccess to mains power or are equipped with high capacity batteries. In addition, MESH_ROUTERstypically have multiple radio interfaces, which significantly increases the transmission capacity ifthe radios are operated on orthogonal channels. In contrast to MESH_ROUTERs, MESH_CLIENTsare relatively resource constrained mobile client devices, such as WiFi-enabled PDAs. Thesedevices usually have only a single radio interface and their key constraint is limited battery power.

A WMN that is entirely comprised of MESH_ROUTERs is referred to as an infrastructure WMN,whereas a client WMN is a network made up of client devices only (Akyildiz and Wang, 2005). Aclient WMN is essentially identical to a pure mobile ad-hoc network (MANET) (Pirzada et al, 2006),and we can therefore consider WMNs a superset of MANETs. A hybrid WMN, such as illustrated inFigure 1, consists of both MESH_ROUTERs and MESH_CLIENTs, with both types of nodesperforming routing and forwarding functionality. In this case, MESH_ROUTERs form the (wireless)backbone of a hybrid WMN, whereas MESH_CLIENTs can be seen as a dynamic extension.

Current routing protocols for WMNs can be roughly grouped into two categories. The firstcategory consists of protocols that are based on traditional routing protocols for wired networkssuch as RIP (Routing Information Protocol) (Malkin, 1998) or OSPF (Open Shortest Path First)(Moy, 1998). Since these protocols are not able to handle node mobility or highly dynamic networksin general, their application is restricted to relatively static infrastructure WMNs.

The second category consists of protocols that are based on MANET routing protocols.Tremendous research efforts have been made in this area over the last few years and an impressivenumber of mobile ad-hoc routing protocols has been proposed (Royer and Toh, 1999). These

Figure 1: Hybrid Wireless Mesh Network




protocols were designed for networks with highly mobile and typically power constrained devices. Asa consequence, they are able to handle node mobility and the generally dynamic nature of clientWMNs and hybrid WMNs, which is why they form the basis of most current WMN routing protocols.

However, since these routing protocols have been designed for relatively homogeneousMANETs, consisting entirely of resource constrained mobile devices, they do not perform optimallyin highly heterogeneous hybrid WMNs. Current WMN routing protocols do not differentiate betweendifferent types of nodes MESH_CLIENTs and MESH_ROUTERs) in the network and are thereforeunable to take full advantage of high capacity MESH_ROUTERs in hybrid WMNs.

A fundamental problem of multi-hop wireless networks in general and WMNs in particular isthe limited scalability and the degradation of performance with increasing path length. Oneapproach to overcome this problem is to use multiple radio interfaces per node, operating onorthogonal channels. Multi-radio nodes have significantly increased capacity, due to reducedinterference and the ability of full-duplex communication, which is not supported by single radionodes. In order to achieve optimal performance in a multi-radio WMN, an efficient channelallocation and selection scheme is required. Even with complete knowledge of the networktopology, optimal channel assignment is very difficult to achieve and is considered an NP-hardproblem (Raniwala and Chiueh, 2005).

The routing protocol presented in this paper aims to achieve two goals. First, it tries to makeoptimal use of high capacity MESH_ROUTERs in a hybrid WMN by routing packets along pathsconsisting of MESH_ROUTERs whenever possible. This not only increases the overall throughputand reduces latency, it also helps to conserve the battery power of client devices. Secondly, wepresent a very simple yet effective scheme that tries to maximize per-path channel diversity. Forexample, the scheme aims to prevent the use of the same channel on neighbouring links of an end-to-end path, which would lead to significant interference and performance degradation.

The Ad-hoc On-demand Distance Vector (AODV) routing protocol (Perkins et al, 2003) formsthe basis of our work. AODV is a popular reactive MANET routing protocol with excellentscalability properties. Even though AODV has inherent support for multi-radio nodes, it lacks built-in support for optimal channel or interface selection and is therefore unable to maximize channeldiversity. AODV has been designed for homogeneous MANETs where nodes have similar computa -tional communication resources and are also similar in terms of their level and pattern of mobility.Thus, AODV will not be able to perform optimally if deployed on a hybrid WMN with highcapacity MESH_ROUTERs that are static and are equipped with multiple radios.

In this paper, we present an extended version of AODV, called AODV-HM (AODV-HybridMesh). The key contributions of our work are:

• We propose a simple modification of AODV’s route discovery mechanism to allow selection ofpaths which maximize the use of MESH_ROUTERs and minimize the involvement ofMESH_CLIENTs. These routes are more stable due to the lower mobility of MESH_ROUTERsand provide lower latency, improved packet delivery rates and a lower control packet overhead.

• We integrate a channel selection scheme into AODV’s route discovery mechanism whichincreases channel diversity of end-to-end paths. This reduces interference and contention andfurther increases the packet delivery rate and reduces latency.

The remainder of the paper is organized as follows: Section 2 discusses relevant related work.The AODV-HM protocol is explained in Section 3. In Section 4 we provide details of our simulationenvironment. Simulation results and their analysis are presented in Section 5 with concludingremarks in Section 6.




2. RELATED WORK2.1 HyacinthHyacinth (Raniwala and Chiueh, 2005) is a multi-channel static wireless mesh network protocolthat uses multiple radios and channels to improve the network performance. It supports a fullydistributed channel assignment algorithm, which can dynamically adapt to varying traffic loads. Ituses a spanning-tree based routing algorithm (IEEE, 2003) to load balance the network as well asto rectify route failures. The MESH_ROUTERs having access to the wired network are consideredas the root nodes of the spanning tree. Hyacinth’s channel assignment algorithm breaks a single-channel collision domain into multiple collision domains, each operating on a different frequency.The channel assignment algorithm operates in two phases: Neighbour-Interface Binding andInterface-Channel Assignment.

In the first phase each node separates its interfaces into UP-NICs and DOWN-NICs. Each nodehas control to change the channel on its DOWN-NICs only. During the second phase each nodeexchanges a periodic message, which contains the channel usage status, with its neighbours in theinterference range. Using the per-channel total load information a node can issue a change channelmessage to its neighbour in order to switch to a least used channel. The advantage of this channelassignment scheme is that a fat-tree architecture is obtained in which links close to the root of thespanning tree are given higher bandwidth. The channel assignment is further integrated with therouting process. Each node having routing information to the root advertises this information toone-hop neighbours. This advertisement also contains the cost metric, which comprises of the hop-count and the residual uplink capacity. Each node receiving this advertisement makes a decision,based upon the cost, as regards to joining the advertising node. If the node decides to join, it sendsan acceptance message to the advertising node and a departing message to the parent node withwhich it was previously attached. New nodes joining the network broadcast HELLO packets, suchas to initiate the joining process by the neighbouring nodes.

2.2 Single-Radio Multi-Channel Routing ProtocolThe Multi-Channel Routing Protocol (MCRP) (So and Vaidya, 2004) is a routing protocol speci -fically designed for networks with single-radio nodes, which can support a channel switching delayof 80 �s or less (Kyasanur and Vaidya 2005). The protocol assigns channels to data flows ratherthan assigning channels to nodes. This implies that all nodes supporting a flow have to be on onecommon channel. The advantage of this mechanism is that once the route is established, nodes arenot required to switch channels for the duration of the flow. The protocol considers all nodes in thenetwork to be in essentially one of four states: free, locked, switching or hard-locked. The freenodes are nodes that are not supporting any flow at the moment. Locked nodes are those that arecurrently supporting one flow. A switching node is one that is supporting two or more flows ondifferent channels. A hard-locked node is one, which due to certain constraints, cannot become aswitching node. MCRP benefits from multiple channels without modifying the MAC protocol.

The routing scheme is similar to that of AODV. However, each Route Request packet (RREQ)is broadcast on each channel in a round robin manner and each receiving node also rebroadcasts theRREQ. Intermediate nodes also create a Reverse Route to the source and maintain a channelnumber for the next hop with each Reverse Route table entry. To convey the channel informationof the next hop, each RREQ contains the operating channel2 number of the hop sending the RREQ.The RREQ also contains the channel table and flow table to be propagated along with each RREQ.

2 The operating channel of a node is the default channel on which it is generally listening on.




The channel table contains the count of similar channels being consecutively used on a single flowpath. The flow tables maintain a count of simultaneous flows being carried out on a single channel.These tables are used by the destination node to make a decision regarding the selection of theoptimal route from multiple received RREQs. The Route Reply packet (RREP) is unicast from thedestination to the source on the optimal path. All nodes forwarding the RREQ change theiroperating channels to the channel selected by the destination.

2.3 Multi-Radio Link Quality Source RoutingThe Multi-Radio Link Quality Source Routing (MR-LQSR) (Draves et al, 2004) protocol has beendeveloped by Microsoft for static community wireless networks. The protocol works in conjunctionwith the Mesh Connectivity Layer (MCL). The MCL permits higher layer applications to connectto the wireless mesh network using WiFi or WiMAX. The MCL implements an interposition layerbetween the link and network layers. It essentially consists of a loadable driver, which acts as avirtual network adapter with the ability to multiplex several physical adapters. Routing of packetsis carried out by MR-LQSR, which is an optimized version of the Dynamic Source Routing (DSR)protocol (Johnson et al, 2003).

The MR-LQSR protocol assumes that the number of wireless interfaces is equal to the numberof channels being used in the network. The protocol identifies all nodes in the wireless meshnetwork and assigns weights to all possible links. To do so, the link information including channelassignment, bandwidth and loss rates are propagated to all nodes in the network. This propagationis combined with the delivery of DSR control packets. The Expected Transmission Time (ETT) oneach link is computed using the Expected Transmission Count (ETX), bandwidth and packet loss.The ETT metric is further used to compute the Weighted Cumulative Expected Transmission Time(WCETT), which defines the path metric designed for multi-radio WMNs. The WCETT is thenapplied to the Link Cache scheme of the DSR protocol. In native DSR, the default cost of links isset to one. Hence, when the Dijkstra algorithm is executed over the link cache by a source node, theshortest path in terms of number of hops is always returned. However, when the WCETT is used asthe link cost, the protocol aims to return the path in terms of link bandwidth, loss rate and channeldiversity.

2.4 Multi-Channel Routing Protocol The Multi-Channel Routing (MCR) protocol (Kyasanur and Vaidya, 2006) has been developed fordynamic WMNs, where nodes have multiple wireless interfaces, each supporting multiple channels.The protocol makes use of an interface switching mechanism to assign interfaces to channels. Twotypes of interfaces are assumed: fixed and switchable. In fixed interfaces, K number of interfacesout of a total M interfaces are assumed to be operating on K fixed channels. In the switchableinterfaces, the remaining interfaces are dynamically assigned to any of the remaining channels.Switching is carried out depending upon the maximum number of data packets queued for a singlechannel. Multiple queues are maintained for all switchable interfaces. Each node maintains aneighbour table and a channel usage list. The neighbour table contains information regarding thefixed channels used by the node’s neighbours. The channel usage list contains the count of nodesthat are using each channel as their fixed channel. Each node periodically transmits a HELLOpacket on all channels, containing the node’s fixed channel number. Each node receiving theHELLO packet updates its neighbour table and channel usage list. The information from the tableand list is used to control the channel and interface switching mechanism. The switchingmechanism assists the MCR protocol in finding routes over multiple channels.




MCR uses a new routing metric, which is computed as a function of channel diversity, interfaceswitching cost and hop-counts. The diversity cost is assigned according to the least number ofchannels used in a route. Thus, a route with a larger number of distinct channels in a route isconsidered to be having a lower diversity cost. The switching cost is used to minimize the frequentswitching of wireless interfaces. The route discovery mechanism of MCR is similar to that of DSR.In addition, each RREQ also contains the channel number and switching cost. When the destinationreceives the RREQ, it computes the diversity cost (number of channels in the RREQ) and theswitching cost (sum of all link switching costs). These costs help the destination in determining andselecting the optimal path available between the source and the destination.

Table 1 presents a comparison of the discussed protocols. All protocols assume homogeneous nodeconfigurations i.e. equal number of interfaces and do not differentiate between different node types.Hyacinth and MR-LQSR have been specifically designed for infrastructure WMNs and have nosupport for client mobility. Hyacinth, MR-LQSR and MCR use interface switching to improve uponthe routing performance in the network, where the interfaces are switched dynamically to differentchannels. MCRP makes use of channel switching on a single interface to connect to nodes operatingon the same channel, to improve upon the network performance. Both interface and channel switchingachieve efficient use of the available spectrum. However, the virtual switching protocol and theconstant switching incurs significant delays causing excessive jitter (Chandra and Bahl, 2004; Draveset al, 2004). In the remainder of this paper, we will present and discuss AODV-HM, which can achieveefficient spectrum usage without the need for a complex and expensive channel switching mechanism.

3. MESH-AWARE AODV PROTOCOL3.1 Standard AODVThe AODV is inherently a distance vector routing protocol that has been optimised for ad-hocwireless networks. It is an on demand or reactive protocol, as it finds the routes only when required.

Protocols Research Year Interface Mobility Derivative Routing RemarksGroup Metric

Hyacinth Stony Brook 2005 Multi-Radio No Spanning Hop Count, • Independent channel University Tree link/path switching protocol

loads required

MCRP University 2004 Single-Radio Yes AODV Hop Count • High speed interfaceof Illinois, switching requiredUrbana- • Complex to handleChampaign multiple flows

MR-LQSR Microsoft 2004 Multi-Radio No DSR WCETT • Proprietary MeshResearch Connectivity Layer

• Bandwidth computed using packet probes

• Requires propagation of intermediary node ETT’sto source node

MCR University 2006 Multi-Radio Yes DSR Channel • Hybrid of fixed andof Illinois, diversity and switchable channels Urbana- switching • Periodic distribution of Champaign cost channel usage lists

Table 1: Comparison of Recent WMN Protocols




AODV borrows basic route establishment and maintenance mechanisms from the DSR protocol,and hop-to-hop routing vectors from the Destination-Sequenced Distance-Vector (DSDV) routingprotocol (Perkins and Bhagwat, 1994). Multi-path support has also been added to AODV through anumber of extensions (Li and Cuthbert, 2004; Marina and Das, 2001), permitting discovery andestablishment of loop-free and disjoint alternate paths. To avoid the problem of routing loops,AODV makes extensive use of sequence numbers in control packets. When a source node intendsto communicate with a destination node whose route is not known, it broadcasts a Route Requestpacket (RREQ). Each RREQ contains an ID, source and destination node IP addresses, andsequence numbers together with a hop count and control flags. The ID field uniquely identifies theRREQ; the sequence numbers indicate the freshness of control packets and the hop-count maintainsthe number of nodes between the source and the destination. Each recipient of the RREQ that hasnot seen the source IP and RREQ ID pair or does not have a fresher (with larger sequence number)route to the destination rebroadcasts the same packet after incrementing the hop-count.

Intermediate nodes create and preserve a Reverse Route to the source node for a certain intervalof time. When the RREQ reaches the destination node or any node that has a fresh route to thedestination, a Route Reply packet (RREP) is generated and unicast back to the source of the RREQ.Each RREP contains the destination sequence number, the source and the destination IP addresses,route lifetime, and a hop count and control flags. Each intermediary node that receives the RREPincrements the hop-count, establishes a Forward Route to the source of the packet, and transmits thepacket via the Reverse Route. To preserve connectivity information, each node executing AODVcan use link-layer feedback or periodic HELLO packets to detect link breakages to nodes that it con -siders as its immediate neighbours. In case a link break is detected for a next hop of an active route,a Route Error packet (RERR) is sent to its active neighbours that were using that particular route.

When using AODV in a network with multi-radio nodes, each RREQ is broadcast on allinterfaces. In order to avoid broadcast storms, a random delay is added in the transmission of eachRREQ. Intermediate nodes with one or more interfaces operating on a shared channel, receive theRREQ and create a Reverse Route that points towards the source node. If the RREQ is a duplicate,it is simply dropped. The first received RREQ received by the destination or any intermediary nodeis selected and all other RREQs are discarded. The RREP is generated in response to the selectedRREQ, and is sent back to the source node on the existing Reverse Route.

3.2 AODV-HMWe have made the following assumptions for the design and evaluation of our AODV-HM protocol:

• All MESH_CLIENTs and MESH_ROUTERs should have at least one common operatingchannel.

• The transmission and reception ranges of the wireless transceivers are comparable.• The wireless antennas are omni-directional.

The aim of AODV-HM is to maximize the involvement of MESH_ROUTERs into the routingprocess without significantly lengthening the paths. In addition, we want to maximize channeldiversity in the selected paths. To implement these features we make two changes to the RREQheader. First, we add a 4-bit counter (MR-Count) indicating the number of MESH_ROUTERsencountered on the path taken by the RREQ. We further add a 7-bit field (Rec-Chan), whichadvertises the optimal channel to be used for the Reverse Route. We have used the existing 11reserved bits in the RREQ header. The first four bits represent the MR-Count while the remainingseven represent the Rec-Chan as shown in Figure 2.

JRPIT 41.1.QXP:Layout 1 13/03/09 12:58 PM 1



3.2.1 Discovery of MESH_ROUTERsWhenever a MESH_CLIENT intends to communicate with another node whose route is notavailable in its routing table, it broadcasts a RREQ on all of its interfaces. Prior to the broadcast, italso sets the MR-Count in the RREQ to zero. Each MESH_ROUTER forwarding the RREQincrements the MR-Count field by one. When the RREQ is received by the destination or anyintermediary node that can respond, the process shown in Figure 3 is initiated.

If the RREQ has not been received earlier3, a RREQ-Timer is started and a RREQ-Counter isinitialized. The RREQ-Timer determines the amount of time a node should wait after receipt of thefirst RREQ and before forwarding the optimal RREQ. The RREQ-Timer helps to evaluate alternatecopies of the same RREQ arriving via different paths. The RREQ-Counter maintains a count ofthese copies. All copies of the RREQ are then buffered until the time when either the RREQ-Timerexpires or the RREQ-Counter reaches a certain threshold.

The optimal values for the RREQ-Timer and Counter are primarily dependent upon the averagenode density. In case the density is high, the RREQ-Counter will reach its threshold value wellwithin the RREQ-Timer. However, in case of a sparse network, the RREQ-Counter may never reachits threshold value before the RREQ-Timer expires. In the standard AODV protocol, the minimumRoute Discovery Latency (RDL), i.e. time between transmission of the first RREQ and the receiptof its corresponding RREP, is

RDL = 2 x np x NODE_TRAVERSAL_TIME

where np is the number of nodes on the path (excluding the source node) taken by the RREP.NODE_TRAVERSAL_TIME is the approximate time taken by a packet to pass through one node.

For lower RREQ-Counter values in high density networks, AODV-HM incurs a RDL similar tothat of the standard AODV. However, in sparse networks with large RREQ-Counter values, AODV-HM may incur a maximum RDL of:

RDL = np x NODE_TRAVERSAL_TIME + np x RREQ-Timer

In order to minimize the RDL, a low value of the RREQ-Counter is maintained or the RREQ-Timer is kept as close to the NODE_TRAVERSAL_TIME as possible.

When the RREQ-Timer expires or the RREQ-Counter threshold is reached, the RREQ, forwhich the routing metric (Hop-Count - MR-Count) is minimal, is selected. This selection is donefrom the set of n RREQs, stored in the RREQ Buffer (RREQ-BUFF), as indicated in Equation 1.

(1)

Figure 2: AODV-HM Route Request Packet Header

3 Determinable through the Source IP and RREQ ID mapping




Figure 3: RREQ Processing in AODV-HM




Let’s consider the scenario shown in Figure 4, where MESH_CLIENT-5 (Source) wants tocommunicate with MESH_CLIENT-45 (Destination). The darker nodes represent theMESH_ROUTERs and the remaining nodes are MESH_CLIENTs. Standard AODV does notdistinguish between MESH_ROUTERs and MESH_CLIENTs. Accordingly, when a routediscovery is initiated from MESH_CLIENT-5 for MESH_CLIENT-45, the first arriving RREQ atMESH_CLIENT-45 establishes the route. The first route, between the source and destination,established using standard AODV is represented as follows:

RSD1 = { 5 → 63 → 59 → 55 → 37 → 45 }

Route RSD1 has a hop-count of five and contains three intermediary MESH_ROUTERs and oneMESH_CLIENT. However, there may be occasions where the first RREQ arriving at the destinationcontains no MESH_ROUTERs, e.g. RSD2 = { 5 → 22 → 14 → 35 → 37 → 45 }. If MESH_CLIENTsoperate on a single channel, as is typically the case, the above scenario would lead to a significantperformance degradation over a route consisting only of MESH_CLIENTs (Li et al, 2001).

In contrast, AODV-HM is able to create Reverse Routes that traverse predominantlyMESH_ROUTERs by delaying RREQs at intermediary nodes and selectively forwarding the oneconsisting mostly of MESH_ROUTERs, instead of the first one to arrive. For example,MESH_ROUTER-60 is likely to receive more than one RREQs originating from MESH_CLIENT-5. In case the first RREQ reaches MESH_ROUTER-60 via MESH_CLIENT-34 (with MR-Count=1), and the RREQ reaches MESH_ROUTER-60 via MESH_ROUTER-64 (with MR-Count=2), AODV-HM would select the latter which contains the smaller number ofMESH_CLIENTs. In standard AODV, MESH_ROUTER-60 would simply forward the first RREQreceived from MESH_CLIENT-34.

Similarly, more than one RREQs are received by the destination MESH_CLIENT-45. Let’sassume five RREQs from MESH_CLIENT-5 have reached MESH_CLIENT-45 and stored in theRREQ-BUFF. The paths taken by the five RREQs are as follows:

Figure 4: Route Development in AODV-HM




RSD1 = { 5 → 63 → 34 → 35 → 37 → 45 }RSD2 = { 5 → 63 → 59 → 55 → 51 → 45 }RSD3 = { 5 → 63 → 59 → 55 → 51 → 20 → 45 }RSD4 = { 5 → 63 → 64 → 60 → 56 → 52 → 45}RSD5 = { 5 → 63 → 64 → 40 → 6 → 7 → 45}

Now using Eq. 1, we get the minimum difference between the Hop-Count and MR-Count forthe above routes as follows: RSD1 : 5 - 1 = 4, RSD2 : 5 - 4 = 1, RSD3 : 6 - 4 = 2, RSD4 : 6 - 5 = 1 andRSD5 : 6 - 2 = 4 respectively. The minimum cost is achieved using the RREQ that arrived via routesRSD2 and RSD4. In case two or more RREQs have the same cost, the first of these to arrive isresponded to and the corresponding route is established.

3.2.2 Channel DiversityIn a wireless network, as the physical medium is shared, nodes have to constantly contend with eachother to gain access to the network. All nodes before making a transmission execute the Carrier SenseMultiple Access with Collision Avoidance (CSMA/CA) protocol to avoid future collisions (IEEE,1997). The effectiveness of the CSMA/CA protocol is influenced by the density, mobility and trafficpattern of the network (Bianchi, 2000). In order to minimize packet collisions and contention, thephysical medium is generally segregated using non-interfering channels. This in turn reduces thenumber of nodes contending per channel, which also lowers the number of packet collisions.

As mentioned earlier, the standard AODV protocol typically adds some random delay prior tothe transmission of RREQs over multiple interfaces. Thus, the Reverse and Forward Routes may ormay not have similar channel assignments4. For example, in Figure 4 the route betweenMESH_CLIENT-5 and MESH_CLIENT-45 indicated with dashed arrows includes threeMESH_ROUTERs: 55, 59 and 63. Let’s assume that the MESH_CLIENTs have one radio eachoperating on Channel 1 (CH-1) and that the MESH_ROUTERs have three radios each, operatingon Channel 1 (CH-1), Channel 6 (CH-6) and Channel 11 (CH-11) respectively. During the routediscovery process, MESH_ROUTER-63 introduces a small random delay before forwarding theRREQ on each of its three channels. If we assume the smallest delay is selected for CH-6,MESH_ROUTER-59 receives the first RREQ via this channel. In this case, the Reverse Route iscreated to MESH_CLIENT-5 via CH-6. When MESH_ROUTER-59 retransmits the RREQ, itemploys a similar mechanism.

However, this time the first RREQ to reach MESH_ROUTER-55 is via CH-1. Thus,MESH_ROUTER-55 creates the Reverse Route to the source MESH_CLIENT-5 via CH-1.Similarly, the Reverse Route from MESH_CLIENT-37 is created via CH-1. This essentiallyintroduces another collision domain between MESH_ROUTER-59 and MESH_CLIENT-37.Packets sent from MESH_ROUTER-59 to MESH_ROUTER-55, as well as packets sent fromMESH_ROUTER-55 to MESH_CLIENT-37 have to contend for the same medium since they allshare the same common channel (CH-1). Thus, the worst case channel selection strategy scenariofor a route traversing through nodes with multiple radios may degrade to that of a path comprisedof single-radio nodes.

In AODV-HM, we use a simple mechanism to achieve effective channel assignment duringroute discovery. Each node, before propagating a RREQ, appends the Recommended Channel (Rec-Chan) to the RREQ, as shown in Figure 2 and Figure 3. The Rec-Channel informs the RREQ

4 The first RREQ received on any interface determines the channel used for the Reverse Route to the source node.




recipient about the desired channel to be used for creating the Reverse Route. The Rec-Chan valueis implemented as a 7 bit number, and its value is set according to Table 2. The first three bits definethe IEEE physical layer standard the radio is operating on. The following 4 bits indicate the specificchannel number. For example, in the network shown in Figure 4, all MESH_CLIENTs are operatinga single radio and are tuned to CH-1 of 802.11b. In this case, Rec-Chan will have a value of 17(0010001).

If a node operates only one radio, the Vacant Channel is the current operating channel. A nodewith multiple radios has the discretion to recommend any Vacant Channel. The Vacant Channel isselected based upon the following two criteria:

• he Rec-Chan is not interfering with the channel being used on the Reverse Route.

• The Rec-Chan is the least loaded channel.

A RREQ can be received by a multi-radio node on one of its Receive Channels (Rx-Chan).However, depending upon the current assignment of channels to the interfaces, the Rx-Chan mayor may not be equal to the Rec-Chan. For example, a node could have all of its radios tuned tochannels other than the Rec-Chan, which would make it impossible create a Reverse Route usingthe Rec-Chan. In case a node has an interface operating on Rec-Chan, it creates the Reverse Routeto the previous hop using that interface, otherwise it creates the Reverse Route via the Rx-Chan.

Coming back to the example of Figure 4, before initiating the RREQ, MESH_CLIENT-5 setsthe Rec-Chan to its current operating channel, i.e. CH-1. As MESH_CLIENT-5 is a single radionode, the RREQ is received by MESH_ROUTER-63 on CH-1 only. In this case Rx-Chan is equalto Rec-Chan, so MESH_ROUTER-63 creates the Reverse Route to MESH_CLIENT-5 using CH-1. MESH_ROUTER-63 is operating on three channels, i.e. CH-1, CH-6 and CH-11. Using thecriteria mentioned above, it now selects the Vacant Channel to be equal to CH-6. The Rec-Chan isthen set to the Vacant Channel and the RREQ is broadcast over all three interfaces.MESH_ROUTER-64, which is also operating the same three channels, now receives three RREQsfrom MESH_ROUTER-63. Since the Rec-Chan in all three RREQs is CH-6, MESH_ROUTER-64creates the Reverse Route to MESH_ROUTER-63 using CH-6.

In our example, MESH_CLIENTs are operating on CH-1 and so this channel would experiencea relatively high load. Thus, MESH_ROUTER-64 selects and recommends the Vacant Channel tobe CH-11, which also does not interfere with the channel used on the Reverse Route. Similarly,MESH_ROUTER-60 creates the Reverse Route via CH-11 and recommends the Vacant ChannelCH-6. In this manner, AODV-HM creates a route between MESH_CLIENT-5 and MESH_CLIENT-45 with less interference and contention compared to the route that is selected by standard AODV.

The routing metric used in AODV-HM aims to minimize the number of MESH_CLIENTspresent in a particular route. This may seem analogous to shortest path routing, but this is not the

IEEE Standard (3-bits) Channel Number(4-bits)

802.11 a 000 1 ~ 16

802.11b 001 1 ~ 16

802.11g 010 1 ~ 16

... ... 1 ~ 16

Table 2: Assignment of Recommended Channels




case, since the metric attempts to route traffic through the MESH_ROUTERs, which in turnmaximize the stability and channel diversity of the routes. A number of other routing metrics likeETX, ETT and WCETT also exist. Of these metrics, only WCETT takes advantage of the channeldiversity, however, its direct application to AODV, which is a distance vector routing protocol,requires extensive modifications to the protocol’s inherent working (Ramachandran et al, 2005).

4. SIMULATION ENVIRONMENTWe evaluated the efficiency of the AODV-HM protocol through extensive simulations in NS-2 (NS,1989), using the Extended Network Simulator (ENS) extensions (Raman and Chebrolu, 2005). AWMN covering an area of 1 square km is established using uniformly distributed staticMESH_ROUTERs and randomly distributed mobile MESH_CLIENTs. Concurrent UDPconnections are established between randomly selected source and destination MESH_CLIENTpairs. A total of four simulations were conducted to evaluate the performance of the AODV-HMprotocol under varying mobility, traffic load and node configurations. The parameters common toall four simulations are listed in Table 3.

The simulations provide the following performance metrics:

Packets Lost: The number of data packets that were lost due to unavailable or incorrect routes,MAC layer collisions or through the saturation of interface queues (Pirzada et al, 2006).

Examined Protocols AODV and AODV-HM

Simulation time 900 seconds

Simulation area 1000 x 1000 m

Propagation model Two-ray Ground Reflection

Mobility model for MESH_CLIENTs Random waypoint

Maximum Speed of MESH CLIENTs5 1 m/s

Transmission range 250 m

Number of Connections5 30

Traffic type CBR (UDP)

Packet Size 128 bytes

Packet Rate 25 pkts/sec

Number of MESH_ROUTERs5 25

Number of MESH_ROUTER Interfaces5 3

Number of MESH_CLIENTs 50

Number of MESH_CLIENT Interfaces 1

MESH_CLIENT RREQ-Counter 5 packets

MESH_ROUTER RREQ-Counter 25 packets

MESH_CLIENT RREQ-Timer 50 ms

MESH_ROUTER RREQ-Timer 250 ms

Table 3: Simulation Parameters

5 The values of these parameters are varied in Simulations 1, 2, 3 and 4 respectively.




Aggregate Goodput: The number of data bits successfully transmitted in the network per second.

Packet Delivery Percentage: The ratio between the number of data packets successfully receivedby destination nodes and the total number of data packets sent by source nodes.

Routing Overhead: The ratio of the total number of control packets generated to the total numberof received data packets.

Average Latency: The mean time in seconds taken by data packets to reach their respectivedestinations.

Path Optimality: The ratio between the length (number of hops) of the shortest possible path andthe actual path taken by data packets.

5. RESULTS AND ANALYSIS5.1 Simulation 1: Varying the MESH_CLIENT SpeedsIn Simulation 1, we have varied the maximum speed of the MESH_CLIENTs from 0 m/s to 20m/s,with increments of 5 m/s. The results, shown in Figure 5, indicate that the packet loss is consistentlylower for AODV-HM compared to standard AODV. This is primarily due to the selection of staticMESH_ROUTERs in the routing process, which offer more stable routes with less contention. Onthe other hand, the standard AODV has no option for prioritizing the routing according to the nodetype. Thus both the MESH_ROUTERs and MESH_CLIENTs are randomly selected in establishinga route. Routes consisting mostly of single-radio MESH_CLIENTs have a higher packet loss due tothe extended contention for the wireless medium, which can lead to saturated interface queues andpackets being dropped. The routes formed by AODV-HM may also involve MESH_CLIENTs in itspaths, but their number is relatively smaller. The lower number of MESH_CLIENTs in the pathmeans improved utilization of the channel diversity and lower contention for the wireless medium.This in effect reduces the packet drop when the AODV-HM protocol is engaged. However, whenthe MESH_CLIENTs move at a higher speed, the routes are frequently broken and recreated. Thuswe see an increase in the number of packets lost with the increase in the network mobility.

The number of packets lost in the network, due to collisions or saturation of interface queues,directly influences the aggregate goodput of the network. AODV-HM shows improved goodputover standard AODV for all speeds. Even though AODV-HM aims to route traffic through theMESH_ROUTERs, at higher speeds the routes become extremely unstable due to the movement ofthe source, destination and intermediary MESH_CLIENTs. The packet delivery rate of AODV-HMranges from 83% at zero mobility to almost 57% at a speed of 20 m/s. Nevertheless, the packetdelivery of AODV-HM is consistently higher than for standard AODV.

AODV-HM has the ability to create more stable routes by preferably involving staticMESH_ROUTERs. This in turn reduces the number of route discoveries in the network, therebylowering the control packet overhead. In addition, as AODV-HM is able to achieve a higher packetdelivery rate, the control packet overhead per received data packet is significantly lower than forAODV. However, it should be noted that AODV-HM does not incur any additional byte overhead,since the MR-Count and Rec-Chan fields occupy existing fields of the AODV RREQ header.

The average latency of the network using AODV-HM is considerably lower than that of thestandard AODV at varying speeds. The lower latency highlights the success of AODV-HM’s routeselection mechanism along with the dynamic channel assignment carried out during the routediscoveries. As mentioned earlier, standard AODV selects the first incoming RREQ. However, thefirst RREQ to arrive does not necessarily arrive via the shortest path (in terms of the number ofhops). Our simulations show that by delaying the RREQs in AODV-HM, the chance of discovering




shorter paths is increased. This is shown in the path optimality metric, which shows that AODV-HM paths have higher path optimality, i.e. they are shorter.

5.2 Simulation 2 : Varying the Traffic LoadIn Simulation 2, we varied the traffic load in the network by increasing the number of simultaneousconnections between the MESH_CLIENTs from 10 to 50, with an increment of 10 connections. Ourresults (Figure 6) show that at lower traffic loads, the performance of AODV-HM is comparable tothat of the standard AODV protocol. However, as the load is increased, the packet loss incurred byAODV increases significantly, thereby decreasing the goodput of the network. The packet deliveryrate for both protocols stays close to 100% up to 20 concurrent connections. Beyond this point, thepacket delivery rate of both protocols starts to degrade. Since the routes created by AODV-HMcontain more MESH_ROUTERs than those created using AODV, we see an improved performanceof the former under increasing traffic loads, relatively to AODV.

The routing packet overhead of AODV-HM also remains lower. The latency of the networkincreases with the increase in the traffic load due to increasing contention for the wireless mediumby nodes operating on interfering channels. However, AODV-HM still manages to maintain asignificant improvement over AODV.

5.3 Simulation 3: Varying the Number of MESH_ROUTERs We have simulated hybrid WMNs with varying numbers of MESH_ROUTERs: 0, 4, 9, 16 and 25.Interestingly, the results (Figure 7) show that even when no MESH_ROUTER is present in the

Figure 5: Results of Simulation 1




network, AODV-HM still has a lower packet loss than standard AODV. This is because AODV-HMdelays the received RREQ and responds to the one with the lowest hop count, since in this case MR-Count=0. In contrast, standard AODV responds to the first RREQ that it received, which may notnecessarily have arrived via the shortest path, due to interference or contention on one of the links.This use of non-shortest paths in AODV increases the total load in the network and increasescontention and packet loss. This is also confirmed by looking at the path optimality of AODV-HM,which is much closer to the shortest possible path6 than the paths created by AODV. Theperformance of both protocols improves with an increasing number of MESH_ROUTERs in thenetwork. However, AODV-HM makes more efficient use of the MESH_ROUTERs and achieves animproved packet delivery rate, decreased packet overhead, and significantly lower latency.

5.4 Simulation 4: Varying the Number of Radios on each MESH_ROUTERIn Simulation 4, we varied the number of radio interfaces in each MESH_ROUTER from 1 to 9,with increments of 2 interfaces. All channels have been configured to be orthogonal and non-interfering with each other. The results of Simulation 4 (Figure 8) reveal that the only time standardAODV outperforms AODV-HM in terms of packet delivery rate is when all nodes are limited to asingle radio operating on the same channel. This means that forcing packets to go viaMESH_ROUTERs, when they do not have a higher capacity than MESH_CLIENTs, can have a


6 The shortest possible path is determined by an omniscient entity present in the NS-2 simulator known as the General Operations Director.




slightly negative impact. In this particular case, MESH_ROUTERs cannot take advantage of thechannel diversity mechanisms of AODV-HM if they are equipped with only a single interface.Nevertheless, since the MESH_ROUTERs are static, they provide more stable routes than mobileMESH_CLIENTs, which results in a reduced packet overhead and lower latency in AODV-HM.The packet delivery rate rapidly improves when the number of interfaces in the MESH_ROUTERsis increased. AODV-HM significantly outperforms standard AODV in these scenarios. However,increasing the number of MESH_ROUTER interfaces to more than three does not show any furtherimprovements. This is due to the fact that three interfaces operating on orthogonal channels aresufficient to provide the required capacity and channel diversity for the network and traffic patternconsidered in our simulation. However, the ideal number of MESH_ROUTER interfaces will varyfor different types of networks with different size, density and traffic load. AODV-HM maintains itssuperior performance over the standard AODV protocol with the increase in the number ofinterfaces. It shows significantly lower packet overhead and latency, and a considerably betterpacket delivery ratio.

6. CONCLUSIONSHybrid WMNs consist of a mix of mobile MESH_CLIENTs and static MESH_ROUTERs. Thesetwo types of node differ considerably in terms of their capacity to forward packets.MESH_ROUTERs are typically much less resource constrained than mobile MESH_CLIENTs, andcan be assumed to be equipped with multiple radio interfaces. Current WMN routing protocols do





not differentiate between the types of node in a WMN, and are therefore not able to exploit theinherent heterogeneity in hybrid WMNs. In this paper, we presented simple extensions to theAODV routing protocol to increase its efficiency in hybrid WMNs. We defined a new routing metricthat allows more efficient use of high capacity MESH_ROUTERs by preferential routing of packetsvia paths traversing the MESH_ROUTERs. In addition, we integrated a channel or interfaceselection scheme to maximize channel diversity and therefore minimize interference on end-to-endpaths. We have performed extensive simulations to evaluate the performance of AODV-HM andcompared it with standard AODV. The results show that AODV-HM consistently outperformsAODV in terms of all our performance metrics and for all simulation scenarios, except for the onespecial case discussed above. Compared to AODV, AODV-HM achieves an increase in the packetdelivery rate of up to 15% in absolute terms, and achieves a reduction in latency by up to 50%.These are encouraging results, given that our proposed changes to the AODV protocol are verysimple and incur only very minor additional overhead or complexity.

ACKNOWLEDGEMENTSNICTA is funded by the Australian Government as represented by the Department of Broadband,Communications and the Digital Economy and the Australian Research Council through the ICTCentre of Excellence program; and the Queensland Government.





REFERENCESAKYILDIZ, I.F. and WANG, X. (2005): A survey on wireless mesh networks. IEEE Communications Magazine, 43; S23-S30.BIANCHI, G. (2000): Performance analysis of the IEEE 802.11 Distributed coordination function. IEEE Journal on Selected

Areas in Communications, 18: 535-547.CHANDRA, R. and BAHL, P. (2004): MultiNet: Connecting to multiple IEEE 802.11 networks using a single wireless card.

Twenty-third Annual Joint Conference of the IEEE Computer and Communications Societies (INFOCOM). IEEE Press,882-893.

DRAVES, R., PADHYE, J. and ZILL , B. (2004): Routing in multi-radio, multi-hop wireless mesh networks. Proceedings ofthe 10th Annual International Conference on Mobile Computing and Networking. ACM Press, 114-128.

IEEE (1997): Wireless LAN medium access control (MAC) and physical layer (PHY) Specifications 802.11.IEEE (2003): Standard for local and metropolitan area networks: Media access control (MAC) Bridges IEEE P802.1D/D41.JOHNSON, D.B., MALTZ, D.A. and HU, Y. (2003): The dynamic source routing protocol for mobile ad hoc networks (DSR).

IETF MANET, Internet Draft.KYASANUR, P. and VAIDYA, N.H. (2005): Routing and interface assignment in multi-channel multi-interface wireless

networks. Proceedings of the IEEE Wireless Communications and Networking Conference. IEEE Press, 2051-2056.KYASANUR, P. and VAIDYA, N.H. (2006): Routing and link-layer protocols for multi-channel multi-interface ad hoc wireless

networks. SIGMOBILE Mobile Computing and Communications Review, 10: 31-43.LI, J., BLAKE, C., DE COUTO, D.S.J., LEE, H.I. and MORRIS, R. (2001): Capacity of ad hoc wireless networks. Proceedings

of the 7th Annual International Conference on Mobile Computing and Networking. ACM Press, 61-69.LI, X. and CUTHBERT, L. (2004): Node-disjointness based multipath routing for mobile ad hoc networks. Proceedings of

the ACM International Workshop on Performance Evaluation of Wireless Ad-hoc, Sensor and Ubiquitous Networks. ACMPress, 23-29.

MALKIN, G. (1998): RIP Version 2. IETF Request for Comments 2453.MARINA, M.K. and DAS, S.R. (2001): On-demand multi path distance vector routing in ad hoc networks. Proceedings of

the Ninth International Conference on Network Protocols (ICNP). IEEE Press, 14-23.MOY, J. (1998): OSPF version 2. IETF Request for Comments 2328.NS 1989: The network simulator. http://www.isi.edu/nsnam/ns/.PERKINS, C., ROYER, E.M. and DAS, S. (2003): Ad hoc on-demand distance vector (AODV) Routing. IETF RFC 3561.PERKINS, C.E. and BHAGWAT, P. (1994): Highly dynamic destination-sequenced distance-vector routing (DSDV) for

mobile computers. Proceedings of the SIGCOMM Conference on Communications, Architectures, Protocols andApplications. ACM Press, 234-244.

PIRZADA, A.A., McDONALD, C. and DATTA, A. (2006): Performance comparison of trust-based reactive routing protocols.IEEE Transactions on Mobile Computing, 5: 695-710.

RAMACHANDRAN, K., BUDDHIKOT, M., CHANDRANMENON, G., MILLER, S., BELDING-ROYER, E. andALMEROTH, K. (2005): On the design and implementation of infrastructure mesh networks. Proceedings of the IEEEWorkshop on Wireless Mesh Networks (WiMesh). IEEE Press, 4-15.

RAMAN, B. and CHEBROLU, C. (2005): Design and evaluation of a new MAC protocol for long-distance 802.11 meshnetworks. Proceedings of the 11th Annual International Conference on Mobile Computing and Networking (MobiCom).ACM Press, 156-169.

RANIWALA, A. and CHIUEH, T.C. (2005): Architecture and algorithms for an IEEE 802.11-based multi-channel wirelessmesh network. Proceedings of the 24th Annual Joint Conference of the IEEE Computer and Communications Societies(INFOCOM). IEEE Press, 2223-2234.

ROYER, E.M. and TOH, C.K. (1999): A review of current routing protocols for ad hoc mobile wireless networks. IEEEPersonal Communications Magazine, 6: 46-55.

SO, J. and VAIDYA, N.H. (2004): A routing protocol for utilizing multiple channels in multi-hop wireless networks with asingle transceiver. Technical Report. Dept. of Computer Science and Coordinated Science Laboratory, University ofIllinois at Urbana-Champaign.

BIOGRAPHICAL NOTESAsad Amir Pirzada is a researcher in the SAFE Networks work-package of theSafeguarding Australia program at NICTA’s Queensland Research Laboratory.He is currently working on Wireless Mesh Networks that can be deployed indisaster scenarios for use by emergency services. He holds a masters degree inComputer Science and a masters degree in Information Security. He receivedhis PhD from the University of Western Australia on Trust and Security issuesin Ad-hoc Wireless Networks. Asad Pirzada’s research interests include datacommunications and network security. Asad Amir Pirzada




Marius Portmann received a PhD in Electrical Engineering from the SwissFederal Institute of Technology (ETH, Zurich) in 2002. He is currently aSenior Lecturer in the School of Information Technology and ElectricalEngineering at The University of Queensland. He is also a Researcher withNICTA’s Queensland Lab where he is mainly working on Wireless MeshNetworks for public safety applications. His general research interests includewireless networks, peer-to-peer systems and network security. He is a memberof IEEE.

Jadwiga Indulska is a Professor in the School of Information Technologyand Electrical Engineering at The University of Queensland. Her researchinterests include pervasive/ubiquitous computing, autonomic networks, mobilecomputing, distributed computing and high speed networks. In the past she ledprojects on mobile and pervasive computing in the Collaborative ResearchCentre on Distributed Systems Technology and currently leads research oncontext-awareness and autonomic networks in NICTA. She is a member of theIEEE Computer Society and the ACM.

Marius Portmann

Jadwiga Indulska


AODV-HM: A Hybrid Mesh Ad-hoc On-demand Distance Vector ...

Documents

Transcript of AODV-HM: A Hybrid Mesh Ad-hoc On-demand Distance Vector ...